Computers & Geosciences 37 (2011) 11741180

Contents lists available at ScienceDirect

Computers & Geosciences

journal homepage: www.elsevier.com/locate/cageo

3D porosity prediction from seismic inversion and neural networks

Emilson Pereira Leite n, Alexandre Campane Vidal
~ Pandia
Calo
geras, 51 CEP, 13083-970 Campinas, SP, Brazil
Department of Geology and Natural Resources, Institute of Geosciences, State University of Campinas, Joao

a r t i c l e i n f o

a b s t r a c t

Article history:
Received 10 March 2010
Received in revised form
19 August 2010
Accepted 31 August 2010
Available online 20 November 2010

In this work, we address the problem of transforming seismic reection data into an intrinsic rock
property model. Specically, we present an application of a methodology that allows interpreters to
obtain effective porosity 3D maps from post-stack 3D seismic amplitude data, using measured density
and sonic well log data as constraints. In this methodology, a 3D acoustic impedance model is calculated
from seismic reection amplitudes by applying an L1-norm sparse-spike inversion algorithm in the time
domain, followed by a recursive inversion performed in the frequency domain. A 3D low-frequency
impedance model is estimated by kriging interpolation of impedance values calculated from well log data.
This low-frequency model is added to the inversion result which otherwise provides only a relative
numerical scale. To convert acoustic impedance into a single reservoir property, a feed-forward Neural
Network (NN) is trained, validated and tested using gamma-ray and acoustic impedance values observed
at the well log positions as input and effective porosity values as target. The trained NN is then applied for
the whole reservoir volume in order to obtain a 3D effective porosity model. While the particular
conclusions drawn from the results obtained in this work cannot be generalized, such results suggest that
this workow can be applied successfully as an aid in reservoir characterization, especially when there is a
strong non-linear relationship between effective porosity and acoustic impedance.
& 2011 Elsevier Ltd. All rights reserved.

Keywords:
Reservoir characterization
Seismic inversion
Feed-forward neural network
Matlab

1. Introduction
During the last decades, several methods for mapping acoustic
impedance from post-stack seismic amplitude data were developed and tested with the aim of providing additional information
for detailed reservoir characterization. Nowadays, most of the
research efforts in this eld are focused in the inversion and
interpretation of variations of seismic reection amplitude with
change in distance between source and receiver (amplitude vs.
offset) from pre-stack data. However, post-stack data obtained
from recorded P-waves are still widely used because of their ready
availability and low time-consuming processing. Because wells in a
reservoir eld are often spaced at hundreds or even thousands of
meters, the ultimate goal of a seismic inversion procedure in the
context of reservoir characterization is to provide models not only
of acoustic impedance but also of other relevant physical properties, such as effective porosity and water saturation, for the interwell regions. Such quantitative interpretations may sometimes
require the use of other seismic attributes additionally to the
traditional seismic reection amplitudes (Rijks and Jauffred, 1991;
Lefeuvre et al., 1995; Russell, 2004; Sancevero et al., 2005;
Soubotcheva, 2006).

Corresponding author. Tel.: + 55 19 35214697; fax: + 55 19 32891097.

E-mail addresses: emilson@ige.unicamp.br, emilson@iag.usp.br (E.P. Leite).

0098-3004/$ - see front matter & 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.cageo.2010.08.001

The seismic inversion method that is presented in this work is

classied as a deterministic inversion method (Russell, 1988).
Although many recent papers have demonstrated some advantages
of geostatistical methods over deterministic methods (Francis,
2005; Robinson, 2001), the latter can still provide geologically
plausible acoustic impedance models at a much lower computational cost. The rst deterministic inversion methods for acoustic
impedance mapping were developed in the late 70 s and became to
known generally as recursive inversion (Lavergne and Willm, 1977;
Lindseth, 1979). The basic premise of those and of all methods that
were subsequently developed in the 1980s is the local validity of
the 1-D convolutional model. During the 1980s, sparse-spike
inversion methods were developed consisting of some techniques
that make use of an additional premise that the reections occur as
sparsely distributed spikes within a layered Earth (Oldenburg et al.,
1983; Russell, 1988). In this case the reectivity function is
mathematically represented as the product of the reection
coefcients and a Dirac delta function shifted by the two-way
travel time to each layer. Two well known methods that fall in this
category are the L1-norm sparse-spike inversion (Sacchi and
Ulrych, 1996), which is applied in the methodology described in
this work, and the maximum likelihood inversion (Hampson and
Russell, 1985).
Prediction of reservoir properties from acoustic impedance can
also be thought as a kind of inversion and traditionally have been
addressed through the application of multivariate statistics and,

E.P. Leite, A.C. Vidal / Computers & Geosciences 37 (2011) 11741180

more recently, Neural Network (NN) methods. The main advantages of NN methods over most traditional statistical methods can
be summarized as follows: (i) the ability to extract nonlinear
relationships between the input data and the target values; (ii) less
sensitivity to the presence of noise in the data; and (iii) there is no
need to known the underlying statistical distribution of the input
data. NN methods have been successfully applied in a wide variety
of applications in reservoir characterization such as porosity and
permeability prediction from seismic and well-log data or seismic
facies/attributes classication (Leiphart and Hart, 2001; Hampson
et al., 2001; Walls et al., 2002; Pramanik et al., 2004; Calderon,
2007). In general, these papers compare performances of NN
models with traditional regression methods, demonstrating that
the former can provide higher correlation coefcient between
actual and predicted reservoir property values and minimize the
problem of sparse well coverage.

2. Methodology
2.1. Seismic inversion
The basic premises behind all seismic inversion methods in the
context of this work are as follows: (i) the Earth can be represented
locally by a stack of plane and parallel layers with constant physical
properties; (ii) the seismic trace s(t) can be represented by the
convolution of the reectivity coefcient series r(t) with a bandlimited wavelet w(t) and the addition of a random noise n(t):
st rtwt nt:

For zero incident angles, r(t) is directly related to the contrast in

the acoustic impedance (AI) of superposed layers through the
expression
rj

IAj 1 IAj
,
IAj 1 IAj

where rj is the reection coefcient at the jth interface of a set of N

superposed layers, and IA rv where r e v are the density and
P-wave velocity, respectively. Under these conditions and assuming that multiple reections were eliminated from the seismic data,
the AI value of each layer can be calculated from the knowledge of
the AI value of the layer above, through a recursive equation


1 rj
IAj 1 IAj
,
3
1rj
which in turn can be generalized to provide the AI value of an
arbitrary M layer by

M 
Y
1 rj
IAM IA1
:
4
1rj
j2
The natural logarithm is applied to both sides of Eq. (4) in order
to obtain a linear approximation:
"
#
M
X
r3 r5
5
lnIAM lnIA1
2 ri i i    ,
3
5
i2
from which we can discard the high-order terms leading to the
expression
AIM AI1 exp2

M
X

rj :

1175

valid for most of the practical cases where rj rj0:3j (Oldenburg

et al., 1983; Berteussen and Ursin, 1983).
In practice, the AI values at the positions of each seismic sample
can be extracted from a 3D model covering the entire seismic
volume, calculated through ordinary kriging of the kwon AI values
at the well log positions. For a properly usage of the recursive
inversion, the seismic traces should be deconvolved into reectivity series as suggested by Eq. (6). To accomplish this, we apply a
constrained sparse-spike optimization procedure that minimizes
the objective function
Jr a

M
X
j1

jrj j

1 1
2
: sWr:
2 s

using the conjugate-gradient algorithm (Shewchuk, 1994). The rst

term in Eq. (7) is provided in order to allow minimization of the L1norm of the reectivities, where a controls the sparsity of the
solution. With the second term, the algorithm also minimizes
the difference between the synthetic seismic traces (Wr) and the
observed traces (s). W is a wavelet coefcient matrix and s is the
standard deviation of the seismic data noise. Other optimization
algorithms can also be used to minimize Eq. (7), such as Iterative

Reweighted Least Squares (Bjorck,

1996) or soft-tresholding algorithms (Loubes and De Geer, 2002).
It is important to point out that this constrained sparse-spike
inversion will provide an impedance model that does not display
the actual reection series but displays only the largest reectors
(Oldenburg et al., 1983). In other words, this means that small
wavelength features in the log impedance curve will not be
recovered by the inversion and, therefore, the interpreter has to
be cautious while analyzing the inversion results.
After estimating r from the seismic amplitudes, then it is
inverted into AI according to the following sequential steps
(Ferguson and Margrave, 1996):
(1) compute the linear trend of a spatial correspondent AI vector
and subtract it, obtaining a residual AIres vector;
(2) compute the Fourier spectra of AIres;
(3) apply Eq. (6) to the reectivity series, obtaining a relative AIrel
vector;
(4) compute the Fourier spectra of AIrel;
(5) determine a scalar a to match the mean power of AIrel and AIres;
(6) multiply the spectra of AIrel by a;
(7) low-pass lter AIres and add to the result of step (6);
(8) inverse Fourier transform the result of step (7); and
(9) add the low-frequency trend from step (1) to the result of
step (8).
It is of course possible to include an extra constraint on
impedances directly in Eq. (7). However, by the approach described
in this paper it is possible to keep control of the frequency contents
involved and the frequency cut-offs to properly add the trend in
acoustic impedance.
Due to the sparse distribution of wells, the low-frequency trend
of step (1) was extracted from spatial correspondent AI traces
estimated by kriging. A low cut-off for coupling the low frequency
trend and a high cut-off were dened by nding where the energy
content of the original seismic traces approaches to zero in the
amplitude spectrum. This characterizes the band-limited nature of
the seismic data.

j2

Eq. (6) is a practical formula used in recursive inversion for

transformation of reectivity into impedance. AI1 is the known
acoustic impedance in the top layer and AIM is that of the Mth layer.
rj is the reection coefcient of the jth layer. This approximation is

2.2. Porosity prediction using Neural Networks

The procedure outlined here can be applied to reservoirs that do
not show a linear relationship between AI and the reservoir property
that needs to be mapped. For the particular example shown in this

1176

E.P. Leite, A.C. Vidal / Computers & Geosciences 37 (2011) 11741180

work, we carried out a NN analysis in order to search for a

relationship between effective porosity (Phie) and other well logs
such as density (RHOB), gamma-ray (GR), and the sonic log (DT). The
employed NN is a three-layer feed-forward system where the
information propagates only in one direction, from the rst to the
third layer. The rst layer contains the input values extracted from
the well logs. The second (hidden) layer consists of an activation
function associated with a set of neurons. These neurons are
represented by weights that are iteratively updated during the
training stage using a gradient descent algorithm. The third layer
outputs the results that are compared with the actual target values at
the end of each training iteration (or epoch) so as to check the meansquared error (MSE) between them.
A hyperbolic tangent sigmoid function (Demuth et al., 2008) is
employed as a transfer function in both the second and the third
layer. For most practical situations there is no deterministic way to
choose the best number of neurons to be used in the second layer
and a trial and error approach has to be applied. Leite and Souza

Fig. 1. 3D seismic data and spatial location of wells. Size of 3D matrix is

301  61  375. In-lines and cross-lines are spaced at about 13 and 27 m, respectively. Time interval is equal to 4 ms.

Filho (2009) presented a more detailed description of the NN

method that was also applied in this work.
As an input for designing a NN model, a sample set obtained from
the well logs is split into training, validation and test subsets. The
training process is carried out until at least one of the following
conditions is met: (i) a minimization of a MSE goal is achieved; (ii)
occurrence of three consecutive non-improvements in the MSE for
the validation subset (early-stopping); or (iii) a maximum number of
iterations are completed. The test subset is used only to estimate the
prediction power of the NN by performing a blind test and it is not
used for building the NN model. The overall workow of the
methodology is shown in Fig. 2.

3. Results and discussion

We carried out a depth-to-time conversion to make the vertical
scale of the well log AI data match the vertical scale of the seismic
data so as to allow an adequate correlation. This conversion was
done using the sonic log and the initial two-way travel time (TWT)
for the rst log sample that provided the highest correlation
coefcient (R) between the synthetic and observed trace. This is
commonly known as a seismic-well tie. The synthetic traces were
calculated using the convolutional model given by Eq. (1). This
requires knowledge of the wavelet representing the seismic pulse.
Thus we performed a deterministic wavelet extraction by writing
Eq. (1) as a linear system and solving for w(t) (Broadhead, 2008).
There are two crucial factors in this procedure that may lead
to poor wavelet estimationincorrect depth-to-time conversion
and incorrect size of the wavelet. In this work, we control the
errors in the former by changing the initial TWT and checking the
value of R. In our tests, by comparing synthetic with observed
traces, we veried that the size of the wavelet should be less than or
equal to 1/5 times the length of the reection coefcient series, so
as to provide the best ts. A common wavelet length was then
determined to be 60 ms for the particular case of this work.
Seismic-well ties were conducted by adjusting ve traces around
each well and retaining the local mean wavelet. Then, a global
mean wavelet was calculated and used for inversion of the traces
away from the wells.

Fig. 2. Flowchart of proposed methodology.

E.P. Leite, A.C. Vidal / Computers & Geosciences 37 (2011) 11741180

An example of this procedure for Well 2 can be visualized

in Fig. 3, where the acoustic impedance log, the estimated
reectivity, the synthetic traces and the observed traces are

1177

shown. Fig. 4 shows the spectral content of the reectivity,

the wavelet, the synthetic traces and the observed traces. The
spectral content is similar for the other four wells in the area and

Fig. 3. Example of seismic-well tie for Well 2: (a) impedance log converted to two-way time and resampled to interval of 4 ms; (b) reectivity obtained after deterministic
wavelet extraction; (c) synthetic traces computed through convolution of reectivity with wavelet and (d) observed traces.

Fig. 4. Normalized amplitude spectrum of (a) reectivity, (b) wavelet, (c) synthetic traces and (d) observed traces near Well 2.

1178

E.P. Leite, A.C. Vidal / Computers & Geosciences 37 (2011) 11741180

low and high cut-offs were dened, after some tests, as 5 and 60 Hz,
respectively.
The reservoir top and base were estimated from well log
markers allowing the denition of minimum and maximum time
values, thus establishing the vertical boundaries of the seismic 3D
grid shown in the subsequent gures. The lateral boundaries were
dened so as to embrace wells that were previously found to have
some oil or gas content in the eld. Fig. 5 shows the AI lowfrequency model obtained by kriging the AI well logs at the wells
depicted in Fig. 1.

Fig. 5. AI low-frequency model obtained by kriging of AI well logs.

In real applications it is commonly difcult to estimate the

standard error of the noise in the seismic data (s in Eq. (7)). A
reasonable assumption is that it corresponds to some percentage of
the peak amplitude of the traces. We used a value of 5% in this work.

Fig. 7. Correlation coefcient between observed seismic traces and synthetic traces
obtained from inverted models around Well 2.

Fig. 6. AI log prole (red curves) vs. inverted AI proles (blue curves) for Well 2 as a varies from 0.02 to 0.035. (For interpretation of the references to color in this gure legend,
the reader is referred to the web version of this article.)

E.P. Leite, A.C. Vidal / Computers & Geosciences 37 (2011) 11741180

The parameter a has to be estimated empirically. We performed

tests by varying a from 0.01 to 1.0 in steps of 0.005 and comparing
the inversion results with the log prole, as well as checking the
value of R between synthetic and observed traces around the wells.
Four results for Well 2 considering a particular range of a, where the
transition of the behavior of the inversion solution is quite noticeable, are presented in Figs. 6 and 7. For a 0.035 the solution is very

Fig. 8. AI obtained through proposed inversion methodology. Color bar is in m/s 

g/cm3. (For interpretation of the references to color in this gure legend, the reader is
referred to the web version of this article.)

1179

smooth, it lacks too much vertical detail and it gives a low R. A value
of a 0.025 was considered to be adequate because the inverted
curve follows the main trends that appear in the log prole while
R is sufciently large. When a is smaller than 0.025, overtting
seems to occur. The behavior of the inverted curve does not change
signicantly outside of the range 0.02 r a r0.035.
The 3D AI inverted model is shown in Fig. 8. A synthetic seismic
model calculated from this inverted model is highly correlated with
the observed seismic data (average R is equal to 0.92). While the
range of AI values is about the same, the inverted model is enriched
in details and can be used for posterior prediction of the reservoir
properties.
In this work we have found that a feed-forward NN can
successfully predict Phie from the joint use of GR and AI logs.
A sample set consisting of 32 log values was extracted from the well
logs for training (60%), validation (20%) and test (20%) of the NN.
The small amount of log values is due to the depth-to-time
conversion, which unavoidably reduces the vertical resolution of
the original well log data. The training must be performed in the
time scale instead of the depth scale to allow posterior prediction in
the entire seismic volume using the inverted AI model as input,
which in turn can only be obtained in the time scale.
In spite of this, the NN models were able to map the test samples
into Phie within an acceptable level of accuracy (R0.84). This was
checked by an iterative cross-validation scheme where the samples
that compose the three subsets were randomly interchanged and a
new NN model was obtained at each iteration. The graphs in Fig. 9
show the results for the NN model that provided the highest overall

Fig. 9. NN training, validation and test. R is correlation coefcient between outputs and actual target Phie values. All values were normalized into range [  1,1].

1180

E.P. Leite, A.C. Vidal / Computers & Geosciences 37 (2011) 11741180

smooth in the vertical direction due to the low vertical resolution of

the seismic data in comparison to the well log data. As a further
step, it would be necessary to test the reservoir performance when
the effective porosity model is used for uid-ow modeling. The
implementation of the inversion algorithm is straightforward and
Matlabs source codes are available from the authors upon request.

Acknowledgement
The authors gratefully acknowledge Petrobras Petroleo Brasileiro SA for the nancial support of this research and for the
development of the project in reservoir characterization.
References
Fig. 10. Phie model obtained by ordinary kriging of Phie well logs. Color bar is in
percentage. (For interpretation of the references to color in this gure legend, the
reader is referred to the web version of this article.)

Fig. 11. Phie model obtained from a designed NN. Color bar is in percentage.
Compared to ordinary kriging, this model shows higher spatial heterogeneity, thus
unraveling reservoir details. (For interpretation of the references to color in this
gure legend, the reader is referred to the web version of this article.)

correlation between the output and target values. Fig. 10 presents a

3D grid calculated by ordinary kriging the Phie values at the well
logs. Having the NN trained model and 3D numerical models of GR
(kriging) and AI (inversion), we were able to estimate a rened 3D
Phie numerical model in the complete seismic volume (Fig. 11).

4. Conclusions
We have presented a two-fold methodology to predict effective
porosity from 3D seismic amplitude data. Seismic amplitude is
rstly converted into acoustic impedance through a constrained
sparse-spike inversion and then this inverted model, together with
a gamma-ray interpolated model, is used to design a Neural
Network capable of predicting effective porosity. For this reservoir
case, the overall correlation coefcient between the Neural Network outputs and the known effective porosity values was equal to
0.84, which is considered high for this kind of application. This also
means that the mist between inverted acoustic impedance and
well log acoustic impedance falls within a tolerable interval.
This methodology can also be used for a quick evaluation of
other reservoir properties such as water saturation, especially
when powerful commercial programs are not available. However, it
is important to notice that while lateral variability can be adequately captured by the inverted model, it tends to be relatively

Berteussen, A.K., Ursin, B., 1983. Approximate computation of the acoustic

impedance from seismic data. Geophysics 41, 882894.

Bjorck,
A., 1996. Numerical Methods for Least Squares Problems, 1st ed. SIAM,
Philadelphia, PA 408pp.
Broadhead, M.K., 2008. The impact of random noise on seismic wavelet estimation.
The Leading Edge 27 (2), 226230.
Calderon, J.E., 2007. Porosity and lithologic estimation using rock physics and multiattribute transforms in Balcon Field, Colombia. Leading Edge 26 (2), 142150.
Demuth, H., Beale, M., Hagan, M., 2008. Neural Network ToolboxTM 6. Users Guide.
The MathWorksTM, Natick, MA 907pp.
Ferguson, R.J., Margrave, G.F., 1996, A simple algorithm for band-limited impedance
inversion: the CREWES research report, vol. 8, pp. 19.
Hampson, D., Russell, B., 1985. Maximum-likelihood seismic inversion. In: Proceedings of the 12th Annual National Canadian Society of Exploration Geophysicists
(CSEG) Meeting. Abstract no. SP-16. Calgary, Alberta.
Hampson, B., Schuelke, J., Quirein, J., 2001. Use of multi-attribute transforms to
predict log properties from seismic data. Geophysics 66 (1), 346.
Lavergne, M., Willm, C., 1977. Inversion of seismograms and pseudo-velocity logs.
Geophysical Prospecting 25, 232250.
Lefeuvre, F.E., Wrolstad, K.H., Zou, K.S., Smith, L.J., Maret, J.-P., Nyein, U.K., 1995.
Sandshale ratio and sandy reservoir properties estimation from seismic
attributes: an integrated study. SEG Expanded Abstracts 95, 108110.
Leiphart, D.J., Hart, B.S., 2001. Comparison of linear regression and a probabilistic neural
network to predict porosity from 3-D seismic attributes in Lower Brushy Canyon
channeled sandstones, southeast New Mexico. Geophysics 66, 13491358.
Leite, E.P., Souza Filho, C.R., 2009. TEXTNNa MATLAB program for textural
classication using neural networks. Computers & Geosciences 35 (10),
20842094.
Lindseth, R.O., 1979. Synthetic sonic logs: a process for stratigraphic interpretation.
Geophysics 44, 326.
Loubes, M.J., De Geer, S.V., 2002. Adaptive estimation with soft thresholding
penalties. Statistica Neerlandica 56 (4), 453478.
Oldenburg, D.W., Scheuer, T., Levy, S., 1983. Recovery of the acoustic impedance
from reection seismograms. Geophysics 48 (10), 13181337.
Pramanik, A.G., Singh, V., Vig, R., Srivastava, K., Tiwary, D.N., 2004. Estimation of
effective porosity using geostatistics and multi-attribute transforms: a case
study. Geophysics 69, 352372.
Rijks, E.J.H., Jauffred, J.C.E.M., 1991. Attribute extraction: an important application in
any detailed 3-D interpretation study. Leading Edge 10 (9), 1119.
Robinson, G., 2001. Stochastic seismic inversion applied to reservoir characterization. CSEG Recorder 26 (1), 3840.
Russell, B.H., 1988. Introduction to Seismic Inversion Methods. Society of Exploration Geophysicists, Tulsa, OK 86pp.
Russell, B.H., 2004. The application of multivariate statistics and neural networks to
the prediction of reservoir parameters using seismic attributes. Ph.D. Dissertation. University of Calgary, Alberta. 392pp.
Sacchi, M.D., Ulrych, T.J., 1996. Bayesian regularization of some seismic operators.
In: Hanson, K.M., Silver, R.N. (Eds.), Maximum Entropy and Bayesian Methods.
Kluwer Academic Publishers, Dordrecht, The Netherlands.
Sancevero, S.S., Remacre, A.Z., Portugal, R.S., Mundim, E.C., 2005. Comparing
deterministic and stochastic seismic inversion for thin-bed reservoir characterization in a turbidite synthetic reference model of Campos Basin, Brazil. Leading
Edge 24 (11), 10941179.
Shewchuk, J.R., 1994. An introduction to the conjugate gradient method without the
agonizing pain. School of Computer Science, Carnegie Mellon University,
Pittsburgh, PA. Report available at /http://www.cs.cmu.edu/  quake-papers/
painless-conjugate-gradient.pdfS.
Soubotcheva, N., 2006. Reservoir Property Prediction from Well-logs, VSP and
Multicomponent Seismic Data: Pikes Peak Heavy Oileld. Saskatchewan. M.Sc.
Thesis. Department of Geology and Geophysics, University of Calgary, Alberta,
94 pp.
Walls, J.D., Taner, M.T., Taylor, G., Smith, M., Carr, M., Derzhi, N., Drummond, J.,
McGuire, D., Morris, S., Bregar, J., Lakings, J., 2002. Seismic reservoir characterization of a U.S. midcontinent uvial system using rock physics, poststack
seismic attributes, and neural networks. The Leading Edge 21, 428436.